University of Southern California Dissertations and Theses

Modeling patient access to point-of-care diagnostic resources in a healthcare small-world network in rural Isaan, Thailand

(USC Thesis Other)

Modeling patient access to point-of-care diagnostic resources in a healthcare small-world network in rural Isaan, Thailand

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content i

MODELING PATIENT ACCESS TO POINT-OF-CARE DIAGNOSTIC RESOURCES IN A
HEALTHCARE SMALL-WORLD NETWORK IN RURAL ISAAN, THAILAND

by

William John Ferguson

A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)

December, 2014

Copyright 2014 William John Ferguson

ii

DEDICATION

I dedicate this thesis to my parents Sara and Tom Ferguson.

My mother because she always told each of her children she loved them the most, and meant it
every time. She shared with me her love of books, which inspired me to be creative and
thoughtful.

My father because through his example I have learned the meaning of real hard work and
dedication.

iii

ACKNOWLEDGMENTS
I would like to acknowledge my thesis advisor Dr. Karen Kemp, thesis committee, and the
faculty at the USC Spatial Sciences Institute who have given me confidence in my spatial
sciences skills. I would also like to acknowledge my mentors Dr. Gerald Kost and Dr. Richard
Louie at UC Davis who have provided such great opportunities over the last four years. I would
like to also acknowledge all my friends and family who have supported me throughout this
process.

iv

TABLE OF CONTENTS
DEDICATION ii
ACKNOWLEDGMENTS iii
LIST OF TABLES vii
LIST OF FIGURES viii
LIST OF ABBREVIATIONS x
ABSTRACT xi
CHAPTER 1: INTRODUCTION 1
1.1 Motivation 2
CHAPTER 2: BACKGROUND AND LITERATURE REVIEW 5
2.1 GIS in Public Health 5
2.2 Social Network Analysis & Physical Network Analysis 6
2.3 Small-world networks and Point-of-Care Technologies 6
2.4 POCT SWN Analysis 8
2.5 Modeling Health Access 9
2.5.1 Raster-Based Approach 9
2.5.2 Vector-Based Approach 10
2.5.3 Comparing Vector- and Raster-Based Approaches 10
2.6 Limitation of Previous Health Access Studies 10
2.7 Summary 11
CHAPTER 3: METHODOLOGY 13
3.1 Study Area 14
3.2 Data 16
v

3.2.1 Population Aggregations 17
3.2.2 Health Resource Facilities 19
3.2.3 Diagnostic Technology Locations 20
3.2.4 Roads 21
3.2.5 Locations of missing data 23
3.3 Methodology 24
3.3.1 Network Creation 25
3.3.1.1 Isolated aggregated population points 26
3.3.2 Sensitivity analyses of travel speeds and integrate tolerance 27
3.3.2.1 Effect of travel speed 28
3.3.2.2 Effect of integrate tolerance 29
3.4 Defining and Improving Access to Cardiac Care 30
3.4.1 Current Health Access 30
3.4.2 Widespread Strategy 30
3.4.3 Resource Limited Strategy 31
CHAPTER 4: RESULTS 32
4.1 Current Access to Diagnostic Technology and Care for Cardiac Events 32
4.2 Health Access to Cardiac Care with Widespread Implementation Strategy 37
4.3 Optimizing Health Access to Cardiac Care 40
CHAPTER 5: DISCUSSION 46
5.1 Data Sources 46
5.1.1 Population 46
5.1.2 Roads 47
vi

5.2 Critically Evaluating the Model 47
5.3 How this model fits into future public health decision 48
CHAPTER 6: CONCLUSIONS 50
REFERENCES 52

vii

LIST OF TABLES
Table 1 Definition of Core Concepts 1
Table 2 Data sources 16
Table 3 OpenStreetMap populated place descriptions 17
Table 4 Study area populated place quantities and estimated populations 18
Table 5 Health resource facilities within each province in study area 20
Table 6 OpenStreetMap road descriptions 22
Table 7 Road travel speed classification 28
Table 8 Results for the travel speed analysis 29
Table 9 Results for the tolerance sensitivity analysis 29
Table 10 Mean (SD) travel time for current health access 32
Table 11 Mean travel time comparison between different policy implementation strategies 38
Table 12 Travel time comparison between different low-resource implementation strategies. 41

viii

LIST OF FIGURES
Figure 1 Visualizes how a small-world network differs from a regular (left) and random (right)
network. (Figure taken from Watts and Strogatz 1998) .................................................... 7
Figure 2 Two-step health access calculation ................................................................................ 14
Figure 3 Location of the study area used in this project as compared to the Isaan region ........... 15
Figure 4 OpenStreetMap aggregated population places ............................................................... 18
Figure 5 Hospitals and health promoting hospitals....................................................................... 20
Figure 6 Current cardiac diagnostic support in the study area ...................................................... 21
Figure 7 Added roads .................................................................................................................... 23
Figure 8 Missing area with low OpenStreetMap data .................................................................. 24
Figure 9 Methods for creating network ........................................................................................ 26
Figure 10 Populated places on isolated road network .................................................................. 27
Figure 11 Selected routes for closest diagnostic facility based on travel time ............................. 33
Figure 12 Selected routes from diagnostic facilities to Srinagarind hospital. .............................. 34
Figure 13 Populated place travel times to nearest diagnosis ........................................................ 35
Figure 14 Populated place travel time to care ............................................................................... 36
Figure 15 Travel time histogram for selected routes of current health access .............................. 37
Figure 16 Comparison of population health access between different policy implementation
strategies ........................................................................................................................... 39
Figure 17 Histogram comparison between different health access policy implementation
strategies ........................................................................................................................... 40
Figure 18 Travel time visual comparison between different low-resource implementation
strategies ........................................................................................................................... 43
ix

Figure 19 Travel time hot spot analysis for visual comparison between different low-resource
implementation strategies ................................................................................................. 44
Figure 20 Histogram comparison between different low-resource implementation strategies .... 45

x

LIST OF ABBREVIATIONS
OSM OpenStreetMaps
POCT Point-of-care Technologies
SNA Social Network Analysis
SD Standard Deviation
SWN Small-world Networks

xi

ABSTRACT
Rapid and accurate diagnoses are important because they drive evidence-based care in health
systems. Point-of-care technologies (POCT) can aid in diagnosis by bringing advanced
technologies out of hospital or clinical settings and closer to the patients. Health networks are
constrained by natural connectivity in the interactions between geography of resources and social
forces. Using a geographic information system (GIS) we can understand how populations utilize
their health networks, visualize their inefficiencies, and model alternatives. This project focuses
on cardiac care resource in rural Isaan, Thailand. A health access model was created using
ArcGIS Network Analyst 10.1 from data representing aggregated population, roads, health
resource facilities, and diagnostic technologies. This model was used to quantify current cardiac
health access and improve upon that access using both widespread and resource limited
strategies. Sensitivity analysis revealed that altering travel speeds of roads has a large effect on
the calculation of health access. Results indicated that having diagnostic technologies closer to
population allowed the streamlining of care paths. The model allowed for comparison of the
effectiveness of the implementation strategies. This model was created to help put the benefit of
adopting POCT within health networks within perspective. Additionally it can help evaluate
these alternatives diagnostic placement strategies as compared to the current health access and
evaluate the relative costs and benefits.

1

CHAPTER 1: INTRODUCTION
The purpose of this project is to demonstrate how a model can quantify access to cardiac care
within a health network and evaluate ways to improve that access. This study focuses on an area
within the northern Isaan region in Thailand to demonstrate this model; however it was created
to be applied to any geographic location. This project focuses on understanding how point-of-
care technologies (POCT) can improve patient access to care within health networks. POCT are
novel technologies used to obtain diagnostic information at or near the patient site of care (Kost,
Tran, and Louie 2002). POCT affords diagnostic information in places where it previously would
not be available, thus streamlining decision-making at the point of need.
This project builds upon the research by Kost et al. (2010) which surveyed health
facilities within Isaan, Thailand for cardiac diagnostic support. This project adds to that research
by using a GIS to understand how different implementation strategies for POCT can improve
health access for a population. Table 1 defines some of the major concepts used in this project.
These are explained further in the literature review (Chapter 2).

Table 1 Definition of Core Concepts
Concept Description
Health Networks

The resources and infrastructure that allow people to
understand and care for their health.
Small-world networks

Represent the natural connectivity by which the
population utilizes their health networks.
Point-of-Care
Technologies
Medical diagnostic testing at or near the patient site of
care. Due to the miniaturization of laboratory tests,
diagnostic information can be obtained in non-
traditional locations.
Health Access Defined as the ability of populations to utilize their
health resources. In this study it mainly refers to the
physical access, as how a person would travel to reach
their health resources.

2

1.1 Motivation
A geographic information system (GIS) can be used to provide a framework to
understand how populations utilize their health networks. Health access studies quantify
population access to health resources, visualize inefficiencies, and model alternative scenarios
(Clark et al. 2012; Coffee et al. 2012; Ranisinghe et al. 2012). These studies are often limited
because they either do not a) identify social pressures of a health system (e.g, cultural health
related decisions) or b) focus on how diagnostic information is obtained.
Health networks are complex systems whose properties are not explained by the sum of
their parts and are often defined by the natural connectivity that arises from their element
interactions (Luke and Stamatakis 2012). Health networks can be thought of as small-world
networks (SWN), which are not completely random or regularly connected systems (Wattz and
Strogatz 1998), and may not be efficient in terms of the urgent health of its populations. This
study shows that a GIS can be used to identify the SWN that exist within health networks, and to
model inefficiencies and help understand how to make them more efficient.
POCT have been implemented in national disaster caches for disaster preparedness
(Curtis et al. 2013), emergency settings for care optimization (Lewandrowski and Lewandrowski
2013), and low resource settings (Garcia et al. 2013; Mabey et al. 2012). In rural settings, POCT
are often the only alternative to conventional laboratory tests. POCT allow populations to access
diagnostic information in places where they previously would not be available.
POCT result from the miniaturization of conventional laboratory diagnostic tests into
portable forms. Many POCT are defined by what is called the Affordable, Sensitive, Specific,
User-friendly, Rapid/robust, Equipment-free, and Deliverable (ASSURED) criteria (Peeling and
Mabey 2010). The most prevalent POCT are glucose meters, which allow users to monitor their
3

diabetes and self-administer treatment. Effectively, the ASSURED qualities allow POCT to be
used in non-laboratory settings, by non-technical staff, and without infrastructure like in-wall
electricity which allows them to be used in a wide range of environments.
Kost et al. (2010) attempted to quantify SWN relationships for cardiac support in health
facilities in Northeastern Isaan, Thailand. Their study revealed isolated regions that did not have
adequate support for their populations. They suggested several placement strategies for POCT
that would improve patient outcomes. While the study was useful for understanding the SWN, it
did not attempt to quantify population access to these diagnostic resources nor model how the
recommended placement strategies would affect this access.
GIS provides the ability to quantify how to effectively implement POCT in health
networks. If implemented correctly POCT have the ability to help streamline decision making.
This can a) improve patient outcomes, b) save resources including money and time, and c)
ensure that the health networks are sufficiently robust for a disaster or emergency event. This
project demonstrates how a GIS can be used to quantify health access and help make decisions
on how to integrate POCT.
The goal of this project is to create, evaluate, and utilize a spatial model that can be used
to improve health networks by quantifying pathways of individuals towards a diagnosis and then
care. This model was evaluated within the study area used in the Kost et al. (2010) study
focusing on the diagnosis and care of individuals with acute myocardial infarction (heart
disease). The goals are:
 To build and evaluate a workable model that defines health access.
 Use the model to define the current health access to cardiac care within a study
area used in the Kost et al. (2010) study.
4

 Use the model to evaluate implementation strategies that improve health access to
cardiac care.
 Define means to evaluate the outcomes of the implementation strategies against
current access.
The following Chapters discuss a) the literature and background that inspired the project,
b) the study area, data, and methods, c) the results, d) critical analysis of the data used and
methods used, and e) how the spatial model can be used for decisions making within health
networks. The end result of this project seeks to a) understand the benefit that POCT can provide
within SWN and b) quantify this benefit such that it can be put into perspective to the technology
integration costs.

5

CHAPTER 2: BACKGROUND AND LITERATURE REVIEW
This chapter provides an overview to several themes relevant to this study. It begins by
describing the role of GIS in public health. It then evaluates previous methods of quantifying
health networks and explores how a GIS can be used to improve upon those methods using SWN
analysis. Additionally, it discusses how POCT are well-suited to be integrated into SWN to
improve health access. The chapter finishes by critically evaluating previous GIS based attempts
to quantify health access.
2.1 GIS in Public Health
Modern GIS applications in public health can be sorted into four categories (Nykiforuk
and Flaman 2011):
a) disease mapping - the compilation and tracking of data on the incidence, prevalence, and
spread of disease
b) risk analysis - which examine the exposure to hazardous environments,
c) community health profiling - the mapping of information pertaining to the health of a
population in a community, and
d) access and planning - which is ability of the community to use health services
This project focuses on using GIS to quantify and improve access to health resources.
Health access can be described as people’s ability to use health services when they are
needed (Cromley and McLafferty 2012). Health access can further be sorted into five dimensions
of access:
a) availability, how the supply of services meets the population needs,
b) accessibility, which deals with physical access, including travel time and cost,
6

c) accommodation, which focuses on how the services are oriented to the consumer
(waiting times, hours of operation),
d) affordability, which has to do with the cost of services, and
e) acceptability, or the patient’s opinion of the services.
This project focuses on the physical accessibility of resource, particularly how the
availability of diagnostic resources can shed light on inadequacies within a health network.
2.2 Social Network Analysis & Physical Network Analysis
Luke and Stamatakis (2012) discussed the complexity of public health systems, meaning
the outcomes of a system are often greater than the sum of its individual parts. The authors point
out that common medical investigations are inadequate because they break down and study
systems by their individual components. Luke and Stamatakis (2012) go on to state that
evaluating a single component will not help understand the system as a whole and suggest
network analysis as a means to support more appropriate research.
Social Network Analysis (SNA) emphasizes the complexity of health systems and
attempts to understand the relationships between actors within a system (Blanchet and James
2012). These methods have become a recent trend in medical system research (Chambers et al.
2012) underscored by the creation of a journal in 2012 entitled Network Modeling Analysis in
Health Informatics and Bioinformatics. However, SNA is inadequate to describe health networks
because the studies often ignore the role of geography in the relationships between people and
their health resources.
2.3 Small-world networks and Point-of-Care Technologies
Watts and Strogatz (1998) argue that conventional network analysis is inadequate
because it assumes the connected topology of biological and social systems are either completely
7

random or completely regular. Instead, they lie somewhere in the middle, and can be considered
a ‘small-world’ as popularized by a 1969 play “Six Degrees of Separation” (Watts and Strogatz
1998). This concept asserts that all people are connected through at the most six social jumps
among friends, colleagues, and acquaintances. SWN are not limited to social connections and
have been used to describe physical networks including electrical, road, and health networks.
SWN are visually defined as an interpolation between regular and random networks.
(Figure 1) A ring lattice with n vertices and k edges per vertex is constructed. In a regular
network, edges connect vertices to their immediate neighbors. In a random network, neighbors
are no longer connected as often and instead long range jumps are made to random vertices. A
SWN lies somewhere between a regular and random network; where most neighbors are
connected with a few long range exceptions. This creates short-cuts that not only affect their
immediate vertices, but neighbors as well. In this project, we do not focus on the mathematical
framework but instead apply a theoretical framework to evaluate the healthcare small-world
phenomenon to a health system.

Figure 1 Visualizes how a small-world network differs from a regular
(left) and random (right) network. (Figure taken from Watts and Strogatz
1998)
8

Healthcare SWN evolve naturally due to “geographic terrain, rural locations of
community hospitals, and fastest routes for ambulances” (Kost, Yu, and Tran 2010, p. 97).
POCT are integral for improving the healthcare SWN because they allow information to be
obtained at the site of care allowing key information to serve people faster. (Kost 2012) The
added benefit of introducing POCT into health networks is that it allows them to be more robust
during emergencies, where diagnostic information can be obtained even with infrastructure
damaged or severed (Kost 2012).
2.4 POCT SWN Analysis
A study was performed to understand how rural emergency departments of low-resource
community hospitals diagnose acute myocardial infarction (Kost et al. 2010). That project sought
to understand whether a SWN existed within the Northern Isaan region in Thailand. Health
professionals were surveyed on cardiac testing equipment services, point-of-care resources, and
health care delivery systems. To understand ambulance travel times between hospitals, regional
topographic maps were presented to emergency medical service personnel who indicated
transportation routes used to regional hospitals.
The study revealed isolated regions with large populations that were far away from
diagnosis and treatment. For care of acute myocardial infarction, patients must be transferred to
regional hospitals, which can be as much as four hours away. Inefficient diagnoses delay the
decisions to route patients to care. The study concluded that having POCT integrated into
different levels of hospital infrastructure (i.e. smaller facilities that are closer to populations),
will speed patient diagnosis and thus efficient transfer to locations of care.
The Kost et al. (2010) study is inadequate for two reasons. The first is because it did not
attempt to understand population access to the diagnostic resources or the role they play to the
9

eventual care of individuals. Ensuring that access is quantified either through travel time or
distance from the place of the individual is important because it ensures proper accounting for
the onset of the illness. The second reason is that while the study offers recommendations for
new POCT resources, it does not offer a means to model their effects. A GIS is used in this study
to model population access to diagnostic resources within the SWN and evaluate alternative
POCT placement schemes.
2.5 Modeling Health Access
Network analysis is used to understand the cost, delivery, and accumulation of resources
between links of given connections (Bolstad 2011). GIS health access studies use three data
sources: population, health resources, and road networks. Population data are often obtained
from country census and aggregated to centroids. Roads are used to represent how people can
travel within the health network. Roads are differentiated from each other based on their
qualities, such as number of lanes, road quality, or type. From those qualities average or
maximum travel speeds are estimated and used to determine distance travel time and distance to
nearest facilities. The methods used to locate the closest facility differ from study to study.
2.5.1 Raster-Based Approach
Many health access studies use raster-based methods (Clark et al. 2012; Coffee et al.
2012; Ranisinghe et al. 2012) where any path within the raster is potentially available for travel.
Each cell is assigned a travel cost, usually based on the type of road that passes through it. If
several roads pass through a single cell, the investigator then decides which better represents that
cell travel cost. A path is selected based on the route that costs the least.
10

2.5.2 Vector-Based Approach
The other type of health access investigation employs a vector-based approach (Brabyn
and Skelly 2002, Schuurman et al. 2006; Owen, Obregon, and Jacobsen 2010). In a vector
approach travel is only allowed along the roads. A travel cost is calculated based on the travel
speeds assigned to the roads. Additional travel costs and elements can be added such as turns,
stoplights, and one-way streets. The final route is selected by which combination of roads costs
the least to travel by.
2.5.3 Comparing Vector- and Raster-Based Approaches
Delamater et al. (2012) discuss the nuances of the two approaches by exploring a single
facility access study using both vector- and raster-based approaches. The main problem with
raster-based studies is that the unique topology of the road networks is lost through the coarse
representation in the raster cell. Additionally the raster based approach proved more sensitive
when travel speeds were altered. The result of the study proved that although both approaches
could provide useful outputs, the vector based method may be more appropriate for representing
access.
2.6 Limitation of Previous Health Access Studies
One of the simplest ways to quantify health access is to determine the straight line, or as
the crow flies, distances. Jordan et al. (2004) compared straight-line distances to drive-time from
urban and rural areas in South West England. The study concluded that although straight-line
distances correlated with health access in urban areas, drive-time is a more accurate measure for
rural areas.
Brabyn and Skelly (2002) performed one of the first investigations of physical access of
the local population to New Zealand hospitals using network analysis. In the study, populations
11

were aggregated to census centroids. Then the closest facility to each centroid was selected along
major road networks. Importantly, the analysis did not evaluate the available diagnostic or care
resources of these hospitals. Although a hierarchy was used to indicate differing levels of
capabilities of health facilities, the study did not classify how individual ailments likely would be
handled within the health networks.
More in-depth studies have been performed on health access to individual ailments. Clark
et al. (2012), Coffee et al. (2012), and Ranisinghe et al. (2012) all studied cardiac resources in
Australia using a raster based approach. Clark et al. (2012) and Coffee et al. (2012) used the
Cardiac Accessibility and Remoteness Index of Australia (Cardiac ARIA) to understand access
to cardiac supplies for the population before and after cardiac events. Cardiac ARIA models
access to key medical services by road ambulance in an acute cardiac event. In contrast,
Ranisinhe et al. (2012) measured access to two different intervention techniques for acute
myocardial infarctions.
These three studies focused on the access to care resources for acute myocardial
infarction. They represent a more focused health access study which looks at an individual
ailment within a health network. The studies did not look into how the diagnosis of acute
myocardial infarction would occur, thus they provide no means to judge the efficacy of that
aspect of the care pathway. Understanding how a condition is first diagnosed and then cared for
will lead to a unique insight which may inform decision making on how to improve the health
network.
2.7 Summary
Previous investigations point to the use of network analysis to accurately models health
access using both physical and social forces of health networks. Health access studies have been
12

used to model and improve health networks but they often do not focus on the role of diagnostic
technology. POCT presents a unique tool because it can be easily implemented in new locations.
Accurately evaluating its role in health access may reveal insights on effectiveness of integration
strategies. This project models the role of POCT within an area in Isaan, Thailand to demonstrate
the benefit of adopting POCT within health networks.
13

CHAPTER 3: METHODOLOGY
This project used a spatial model to define a population’s access to cardiac support and how that
access can be improved through POCT. The model defines health access by quantifying how an
individual would travel from a place of origin to a location of diagnosis and then to a location of
care. This process involves four data types: a) populated places, which define where individuals
originate from, b) roads, which define how an individual would travel to locations of diagnosis
and care, c) health resource facilities, which define the locations of diagnosis and care, and d)
cardiac diagnostic resources, or how an individual would obtain a diagnosis for acute myocardial
infarction.
This model used two steps to calculate health access, which are summarized in Figure 2.
The first step involves calculating the travel time along the road network from populated places
to nearest diagnostic facilities. The second step calculates the travel time from the diagnostic
facility to nearest location for cardiac care, in this case Srinagarind Hospital. Srinagarind
Hospital was selected because it contains the only available cardiac care support in the region.
To estimate total health access the travel time from population center to diagnosis is added to the
travel time from diagnosis to Srinagarind hospital.
This study first performed a sensitivity analysis to understand how variations in the
modeling method affect the calculation of health access. Second, it quantified the current health
access according to the resources surveyed by the Kost et al. (2010) study. Third, to improve
health access the study evaluated two themes of POCT implementation strategies. The first
implementation strategy looks into how a widespread POCT integration would affect health
access, the second looks into how a limited resource strategy would affect health access.

14

Figure 2 Two-step health access calculation

The following methods section discusses a) the study area used in this investigation, b)
the data sources, c) network creation and sensitivity analysis, d) current health access calculation,
and e) new implementation strategies.
3.1 Study Area
Thailand was selected as the study area which was previously investigated using paper-
based surveys by the researchers in the Kost et al. 2010 study. Thailand, a member of the
Association of Southeast Asian Nations (ASEAN), borders Cambodia, Malaysia, Myanmar and
Laos. Thailand has a population of approximately seventy million and represents about 11
percent of the total population of ASEAN countries. Thailand consists of seventy seven
provinces. The Isaan region, often referred to simply as the Northeast region of the country, had
21,305,000 people in 2010 and is considered to contain the country’s most rural and poorest
areas.
15

Figure 3 shows the Isaan region which consists of twenty provinces, eight of which are
used in this project. The provinces of Bueng Kan, Sakhon Nakhon, and Nakhon Phanom make
up the main study area, where individuals will originate from. The provinces of Nong Khai,
Udon Thani, Kalasin, and Mukdahan border the main study area and are included because
individuals must travel through these provinces to reach Khon Kaen. Khon Kaen province is the
second largest province in Isaan and has one of the larger cities in the region containing
Srinagarind Hospital, the only location of cardiac care in the region. Additionally these
provinces may contain health resource facilities that are closer to populated places than the ones
in the main study area. The other twelve provinces are not used in this study.

Figure 3 Location of the study area used in this project as compared to the Isaan region
The study area consists of Bueng Kan, which has a size of 4,305 square miles and a
population of 362,754; Sakhon Nakhon, a size of 9,606 square miles and a population of
16

941,810; and Nakhon Phanom, a size of 5,513 square miles and a population of 583,726. Bueng
Kan came into existence in 2011 after it was separated from its neighbor Nong Khai (Law 2014).
3.2 Data
Four data sources were used to represent roads, health resource facilities, diagnostic
technology locations, political boundaries, and populated places within the study area. Table 2
provides an overview of these data and their sources. Before analysis, all data was projected to
the WGS 1984, UTM 47N, which is suitable for use for most of Southeast Asian countries
including Thailand.
Table 2 Data sources
Data Name Theme/Topic Description Date Obtained
OpenStreetMap
http://www.openstreetmap.org/

Roads and
populated
places
User submitted open data
that emphasizes local
knowledge. Contains
both line and point data
with attributes.
Downloaded
May 29th, 2014
MapMagic 2013: Thailand
http://www.thinknet.co.th/
Health
resource
facilities

Proprietary point data
collected and sold by a
Thai based company.
Assumed to
represent the
facilities that
exist as of 2013.
Global Administrative
Areas
http://www.gadm.org/
Thailand
province and
amphoe
boundaries
Lines representing
political boundaries.
Note: Bueng Kan was
established in 2011 and
the new boundary was
added manually.
Downloaded
May 29
th
, 2014
Kost et al. 2010
Diagnostic
technology
locations

Resources for available
cardiac diagnostic
resource were surveyed
in Isaan Region. Contains
attributed point data.
Collected
throughout 2009
and 2010.

17

3.2.1 Population Aggregations
The model used population aggregation points, henceforth referred to as populated
places, to represent where individuals may originate when they have a cardiac event. Since data
about the distribution of population in the study area was not otherwise available,
OpenStreetMap (OSM) data on populated places was used. OSM places represent populated
settlements including cities, towns, villages, and suburbs. Places found within the main study are
described according to OSM definition in Table 3, along with estimated populations, and number
found within the study area. Although the OSM places are most likely not all the population
centers in all provinces, they are assumed, for this demonstration, to be a sufficient
representation of the population distribution in the provinces.

Table 3 OpenStreetMap populated place descriptions
(Descriptions modified from OpenStreetMap places metadata)
Type Description Population Estimate Count
City

Largest urban settlement in
province.
Usually has more than
100,000 people
1
Suburb

A distinct section of an urban
settlement.
Unknown population 1
Town

A second tier urban settlement of
local importance.
More than 10,000 5
Village

A smaller distinct settlement,
smaller than a town.
Less than 10,000 1303
Hamlet A smaller rural community 100 to 200 people 83

Populated places are summarized by province and type in Table 4 and illustrated in
Figure 4. For this study, a place’s population was estimated according to OSM’s descriptions.
Total estimate population was 82 percent of the 2010 Thai Census population. Province
estimates were 95 percent of 2010 Thai Census for Bueng Kan, 81 percent for Sakhon Nakhon,
and 75 percent for Nakhon Phanom.
18

Table 4 Study area populated place quantities and estimated populations
Bueng Kan Sakhon Nakhon Nakhon Phanom
Type
(Population
Estimate)

Quantity

Estimated
Population

Quantity

Estimated
Population

Quantity

Estimated
Population
City
(100,000)
0 0 1 100,000 0 0
Suburb
(100,000)
0 0 1 100,000 0 0
Town
(20,000)
2 40,000 1 20,000 2 40,000
Hamlet
(200)
17 34,000 27 5,400 39 7,800
Village
(1,500)
271 271,000 641 641,000 391 391,000
Total 290 345,000 671 766,400 432 438,800

Figure 4 OpenStreetMap aggregated population places
19

3.2.2 Health Resource Facilities
Health resource facilities were obtained from MapMagic 13, a product developed by
THiNKNET (www.thinknet.co.th), a company based in Thailand. MapMagic 13 is a mapping
application that allows users to visualize and analyze marketing, logistics, and business
geographic data in all seventy-seven Thailand provinces. MapMagic 13 has two designations for
health resource facilities: hospitals, which tend to be larger facilities with more robust diagnostic
and care capabilities, and health promotion hospitals, which represent smaller hospitals including
dentist offices and clinics. MapMagic does not provide much detail on the designations;
examining the health promotion hospitals in more detail reveals that they include dentist offices
and private clinics which may not be appropriate for POCT implementation for public health
access. MapMagic 13 data is not made available as digital files, so these locations were
manually digitized from screen views of their maps. To do this, latitude and longitude
coordinates were recorded to six decimal degrees and loaded into ArcGIS as points.
To eliminate edge effects, or the fact that people may travel to adjacent health facilities in
neighboring provinces, health resource facilities were also digitized for the western and southern
bordering provinces of Nong Khai, Udon Thani, Kalasin, and Mukdahan. This study assumes
that people do not travel to out-of-country resources, so it was not necessary to include data
across the northern and eastern borders in Laos.
Figure 5 shows the locations of health resource facilities that could serve as location for
diagnostic technology, including health resource facilities in the surrounding provinces. Figure 6
also shows the location of Srinagarind Hospital in comparison to diagnostic resources surveyed
currently available. Provincial totals of health facilities are summarized in Table 5. The majority
20

of hospitals and health promoting hospitals fell within Sakhon Nakhon which correlates well
with the region’s larger size and population.

Figure 5 Hospitals and health promoting hospitals

Table 5 Health resource facilities within each province in study area
Province Bueng Kan
Sakhon
Nakhon
Nakhon
Phanom
Hospitals 8 23 13
Health Promoting Hospitals 51 171 125
Total Health Resource Facilities 59 194 138

3.2.3 Diagnostic Technology Locations
Data on the location of point-of-care technology within the Isaan region was collected by
Kost et al. (2010) in 2009 and 2010, who surveyed health facilities within the area. These
21

locations may not represent current POCT status within the provinces; however the locations can
be used to demonstrate the effects for the implementation strategies recommended in that study.

Figure 6 Current cardiac diagnostic support in the study area

3.2.4 Roads
Road data were obtained from OSM. In the OSM data, there are over 26 different
categories of roads. Many of these roads are very specialized, such as link roads which are
smaller roads which connect lanes of highway to other highways or adjacent roads. Since this
study is limited to travel by car over larger distances, it only includes the road types that allow
that form of travel. Table 5 lists the road types used in this study area and their descriptions
22

according to OSM. Roads were obtained for the main area as well as for the provinces of Nong
Khai, Udon Thani, Kalasin, Mukdahan, and Khon Kaen.
The road network for the study area is illustrated in Figure 7. Major roads are highlighted
to indicate faster routes. To connect the health resource facilities to the road network 2,331 roads
were added accounting for 738.3 miles of new roads. The longest road added was 6.3 miles, and
an average of 0.32 miles per segment.

Table 6 OpenStreetMap road descriptions
(Descriptions modified from OpenStreetMap road metadata)
Highway
Type OpenStreetMap Description
Motorway A restricted access major divided highway, normally with two or more
running lanes plus emergency hard shoulder. Equivalent to the
Freeway, Autobahn, etc.
Trunk The most important roads in a country's system that aren't motorways.
Primary The next most important roads in a country's system. (Often link larger
towns.)
Secondary The next most important roads in a country's system. (Often link
smaller towns and villages.)
Tertiary The next most important roads in a country's system.
Unclassified The least most important through roads in a country's system - i.e.
minor roads of a lower classification than tertiary, but which serve a
purpose other than access to properties.
Residential Roads which are primarily lined with and serve as an access to
housing.
Service For access roads to, or within an industrial estate, camp site, business
park, car park etc.
Living
Street
Residential streets where pedestrians have legal priority over cars,
speeds are kept very low and children are allowed to play on the street.
Road A road where the mapper is unable to ascertain the classification from
the information available.
Link Roads The link roads allow movement between roads.

23

Figure 7 Added roads

3.2.5 Locations of missing data
One area in particular had a suspicious lack of roads or populated places in the OSM
data. Figure 8 shows the road network in this area. This omission will impact the travel time
estimates. However, for the purposes of this demonstration, it was concluded that the richness of
the dataset elsewhere in the region is sufficient.

24

3.3 Methodology
The following section describes the method for a) creating a network, b) performing a
sensitivity analysis using different integrate tolerances and road travel speeds, c) understanding
the current health access, d) understanding the health access under widespread integration
policies, and e) using a location-allocation algorithm to select best implementation sites for new
POCT locations under limited resource strategies.

Figure 8 Missing area with low OpenStreetMap data
25

3.3.1 Network Creation
A GIS network is an advanced connectivity model that can be used to represent and study
complex scenarios such as transportation networks. Networks are comprised of: edges (lines),
which represent how entities move along the environment; junctions (points), which dictate how
entities travel from line to line; and turns, which are optional elements which limit the movement
at junctions between edges.
This network is used to represent the means by which people will travel from their
locations to diagnostic support and then care. It was necessary to ensure that the point locations
for Health facilities and population centers fall on the network, otherwise travel between points
could not be modeled. When these points did not fall directly on a road, it was necessary to add
connector roads. Since there was no additional information available to determine where these
roads actually are, these added road segments were created simply as straight lines from the
points to the nearest locations along the roads.
Since the OSM dataset is not provided in a network structure, it was necessary to process
the data to ensure connectivity. Road elements that do not directly connect could vastly alter
results. To ensure that these connections are correct, ArcGIS provides an Integrate tool, which
snaps elements together with a given distance tolerance. The problem with this is choosing the
correct tolerance and understanding the effect of that tolerance on the solutions generated. An
analysis was performed to understand the effect of this tolerance on the network and is described
in Section 3.3.2.2.
Figure 9 illustrates the steps to produce the network. First, connecter roads were added
using the lines from the health resource facilities and populated places to the nearest point along
26

the OSM roads. Next, using the ArcGIS integrate tool, nearby lines were snapped together using
a tolerance distance of one-, five-, and ten-meters.

Figure 9 Methods for creating network
No turns were modeled along this network, meaning no cost was added to the routes
generated from moving to one element to another and that it is possible to turn in any direction at
any intersection. Additionally, interaction between elements could occur at any vertex, or
wherever two elements touched.
3.3.1.1 Isolated aggregated population points
Five aggregated population points were on locations in the network that were isolated
from the rest. Figure 10 displays the location of two of the places illustrating their isolation from
the road network. Only one aggregated population point was on a portion of a road network with
27

a health resource facility, however that health resource facility did not have a means to connect
to Srinagarind Hospital. This was discovered late in the project and deemed a small portion of
the total populated places, thus all five were removed from the analysis.

Figure 10 Populated places on isolated road network

3.3.2 Sensitivity analyses of travel speeds and integrate tolerance
Two sensitivity analyses were performed to understand the effect of travel speed in miles
per hour (MPH) and integrate tolerance on health access. An analysis independently altering
travel speeds (Table 7) and integrate tolerances (one, five, and ten meters) was used to
understand their effect on the models calculation of health access. The sensitivity analysis used
all health resource facilities (both hospitals and health promoting hospitals) to calculate health
access.
28

Table 7 Road travel speed classification

Road Type
Estimated
Guess
(MPH)
Faster
Highways
(MPH)
Slower
Highways
(MPH)
Faster
Streets
(MPH)
Fast Highways,
Fast Streets
(MPH)
Trunk 65 80 55 65 75
Primary 55 70 50 55 75
Secondary 45 60 45 45 75
Tertiary 30 55 40 45 75
Unclassified 25 25 25 45 45
Residential 25 25 25 45 45
Service 15 15 15 45 45
Roads 35 35 35 45 45
Added 35 35 35 45 45
Living Street 25 25 25 45 45

3.3.2.1 Effect of travel speed
The results of the travel speed sensitivity analysis are summarized in Table 8. Mean and
median travel time and distance are shown in the table. Also shown are the standard deviation
(SD), minimum (Min), and maximum (Max) travel time and distances.
The estimated guess used in this analysis resulted in an average travel time to care of 160
minutes. Increasing highway speeds resulted in decrease in travel time by almost 25 percent.
Decreasing highway speeds resulted in no change. Increasing street speeds resulted in a decrease
in travel time by 13.2 percent. Increasing both street and highway speeds resulted in decrease of
36.3 percent, a difference of 62.2 minutes. Travel distance on the other hand was minimally
altered, with a maximum average change of 2.3 percent.
29

Table 8 Results for the travel speed analysis

Estimated
Guess
Faster
Highways
Slower
Highways
Faster
Streets
Fast
Highways
and Streets
Travel
Time
(minutes)

Mean MPH
(SD)
164.0
(38.1)
121.4
(25.8)
164.1
(35.1)
138.9
(32.3)
101.9
(20.7)
Median MPH
(Max/Min)
170.1
(230.5/
70.1)
124.5
(169.2/
57.0)
167.9
(227.6/
75.9)
142.1
(201.4/
60.7)
103
(141.9/
50.8)
Mean Change
Compared to
Educated Guess
N/A 26.0% 0.0% 15.3% 37.9%
Travel
Distance
(miles)
Mean MPH
(SD)
176.8
(30.9)
174.4
(29.1)
172.7
(29.2)
172.9
(29.0)
172.9
(27.6)
Median MPH
(Max/Min)
179.3
(249.9/
107.7)
176.3
(244.7/
106.3)
172.9
(243.7/
107.9)
172.4
(244.0/
106.1)
173.4
(233.3/
105.8)
Mean Change
Compared to
Educated Guess
N/A 1.3% 2.3% 2.2% 2.2%

3.3.2.2 Effect of integrate tolerance
The results of the integrate tolerance sensitivity analysis is summarized in Table 9. The
differences between integrate tolerance thresholds amounted to about one minute and one mile
difference between the average travel time and travel distance respectively. The results indicate
that tolerance thresholds do not have a large effect on the analyses, thus a ten meter tolerance
was used.

Table 9 Results for the tolerance sensitivity analysis
Integrate
Tolerance
Mean (SD) Travel
Time (minutes)
Mean (SD) Travel
Distance (miles)
One meter 165.1 (38.3) 177.7 (23.8)
Five meters 165.0 (38.3) 177.7 (31.6)
Ten meter 163.9 (38.3) 176.7 (31.3)

30

3.4 Defining and Improving Access to Cardiac Care
In this study health access is quantified by travel times and distances from a) populated
places to diagnosis, b) diagnosis to care, and c) the addition of populated places to diagnosis and
diagnosis to care (referred to as just travel from populated places to care).
This section describes how the project quantified the current health access, and then by
using widespread and resource limited implementation strategies, improved upon this access by
adding POCT at strategic locations.
3.4.1 Current Health Access
The first step in improving health access is to define the current access. This was done
using the road networks, road speed, populated places, health resource facilities, and POCT
defined in Section 3.2 and 3.3 to evaluate current access routes and travel times from a)
populated places to diagnosis and b) populated places to care were visualized and compared.
3.4.2 Widespread Strategy
Two analyses were performed to evaluate how implementing diagnostic technologies at
different administrative levels will affect health access. This simulates if the health network
adopted a policy that dictated certain facilities had to implement POCT.
Two scales of integration were used. The first defined health access if POCT existed in
every hospital. The second analysis defined health access if POCT existed in every hospital and
health promoting hospital. For each analysis, health access was quantified and compared to the
current health access defined in Section 3.4.1.
31

3.4.3 Resource Limited Strategy
The resource limited strategy simulates if a health network only has limited resources to
improve their health networks. This analysis can be used to understand how to best utilize
existing resources to improve the population’s health access. It involves understanding how to
rearrange existing resources for better outcomes.
Five analyses were performed which evaluated different resource limited implementation
strategies:
1. Rearranged existing POCT.
2. Added 5 additional POCT while keeping the existing resources where they are.
3. Added 10 additional POCT while keeping the existing resources where they are.
4. Added 5 additional POCT while allowing the existing resources to be rearranged.
5. Added 10 additional POCT while allowing the existing resources to be rearranged.
A location-allocation analysis, using population as a weight, was used to determine the
best locations for the POCT diagnostic resources in each analysis. After the health resource
facilities were selected health access in each implementation strategy was compared to each
other and results obtained in Section 3.4.1.
To visualize the differences between the resource limited implementation strategy results,
a fixed distance euclidean Getis-Ord Gi* hot spot statistic was used. The Getis-Ord GI* statistic
identifies clusters of points with value higher and lower in magnitude than you would expect by
random chance. The magnitudes, in this case represented by the addition of travel time from
populated places to diagnosis and the travel time from diagnosis to care, are compared at
confidence levels of 90 percent (z-score of 1.65), 95 percent (z-score of 1.96), and 99 percent (z-
score of 2.58) to visualize localized hot spots.
32

CHAPTER 4: RESULTS
This chapter discusses the results of the analysis of a) the current health access for cardiac care,
b) the effects of the two widespread implementation strategies of integrating POCT diagnostic
technologies, and c) the effects of resource limited implementation strategies of POCT
diagnostic technologies.
4.1 Current Access to Diagnostic Technology and Care for Cardiac Events
Health access was calculated by determining the quickest route from a) populated places
to the nearest health resource facility with cardiac diagnostic resources (see Figure 6) b)
diagnosis to care, and c) populated places to care. The results are outlined in Table 10 and
contain the mean (SD) travel time and distance.
The analysis splits the total travel time and distance into its health access components: a)
populated places to diagnosis and b) diagnostic resource to care. The average overall travel time
from populated places to care of 3.6 hours is mainly dominated by the travel from diagnosis to
care of 2.8 hours. The minimum travel time was 1.6 hours with a maximum of 5.5 hours, a
difference of 3.9 hours.
Table 10 Mean (SD) travel time for current health access

Health Access
Mean (SD)
Travel Time (minutes)
Populated places to diagnosis 49.9 (25.5)
Diagnosis to care 168.6 (37.1)
Populated places to care 218.5 (39.5)

Figure 11 demonstrates the routes determined by the analysis from populated places to
diagnosis. For populated places on the western edges of Bueng Kan and Sakhon Nakhon the
33

closest diagnosis was out of the provinces. Individuals from these locations must travel west and
in some cases north to reach a diagnosis.

Figure 11 Selected routes for closest diagnostic facility based on travel time

Figure 12 builds on Figure 11 by adding the routes selected from diagnosis to care.
Although at first it seems there is a problem with the routes selected with many heading west
first and then south to Srinagarind Hospital, this reflects the major road that provides
transportation east to west in the middle of Sakhon Nakhon (see Figure 7). For populated places
in the western edges of Sakhon Nakhon, individuals who were identified in Figure 11 to travel
west and north must now backtrack across their paths to reach care.
34

Figure 12 Selected routes from diagnostic facilities to Srinagarind hospital.

Travel time from populated places to diagnosis is indicated in Figure 13 to demonstrate
the dispersion of travel times. Lower travel times occur in the eastern parts of the provinces,
nearer to where the diagnostic facilities are located. Concentrations of higher travel times
occurred in the north-western areas of Sakhon Nakhon and western areas of Bueng Kan where
individuals must travel over 1.5 hours to reach a diagnosis.
35

Figure 13 Populated place travel times to nearest diagnosis

Figure 14 shows the travel time from populated places to care. In the western areas of
Bueng Kan there are groupings of populated places that are under two hours of care.
Immediately surrounding these locations, travel time of these populated places increased to
between 5 and 6 hours. These times correlate with the travel patterns identified in Figure 11 and
12, where individuals in the either travel west or directly east to reach diagnosis. Individuals who
travel east must then backtrack to reach care. Although the model correctly indicated the closer
facility for individuals to travel to for a diagnosis, it may be better to have a longer time to
diagnosis in order to decrease overall time to care.
The longest travel times to care mostly fall in Bueng Kan. A small localized area of
higher travel times can also be located in northern Sakhon Nakhon. This may be because
36

individuals from this area must travel 1.5 to 2 hours to reach a diagnosis and then backtrack to
reach care. The locations further away from Srinagarind Hospital, where you would expect
longer travel times, particularly eastern Nakhon Phanom, have relatively lower travel times.

Figure 14 Populated place travel time to care

Figure 15 shows the histogram for the travel times from populated places to care and
provides a means to understand the distribution of selected routes. Bins were allocated to
represent 15 minute intervals. The frequency is the number of routes which travel times fall
within the 15 minute bins.
37

Figure 15 Travel time histogram for selected routes of current health access

4.2 Health Access to Cardiac Care with Widespread Implementation Strategy
This section identifies the travel routes determined after implementing POCT according
to two widespread implementation strategies. The first strategy integrates POCT within every
hospital. The second strategy integrates POCT within every health resource facility.
Mean travel time for the two strategies are shown in Table 11 compared to results from
Section 4.1. Both implementation strategies offered greatly decreased travel time when
compared to the current diagnostic resource access. The quickest routes were generated when
POCT was implemented in every health resource facility; however it offered little advantage
over integrating POCT in every hospital. The difference between average travel times from
populated places to diagnosis between the two strategies was 9.3 minutes which is negligible
when compared to the travel time from populated places to care. The difference between these
two strategies would only result in a 2.6 percent decrease, potentially a reason to only implement
POCT within hospitals instead of every health resource facility.
38

Table 11 Mean travel time comparison between different policy implementation strategies
Health Access

Current Access

All Hospitals
All Health
Resource Facilities
Average Travel Time
from Pop to Diagnosis
(minutes)
49.9
(25.5)
13.9
(7.4)
4.6
(3.5)
Percent Decrease Over
Current Access
N/A 72.1% 90.8%
Average Travel Time
from Diagnosis to Care
(minutes)
168.6
(37.1)
155.9
(37.9)
159.4
(37.9)
Percent Decrease over
Current Access
N/A 7.5% 5.5%
Total Travel Time
(minutes)
218.5
(39.5)
169.8
(39.9)
164
(38.0)
Mean Travel Time
Percent Decrease Over
Current Access
N/A 22.3% 24.9%

Figure 16 displays the results for the two strategies. In both cases analyses created a
pattern of health access that starts in western Sakhon Nakhon with travel times spreading
smoothly outward with increasing magnitude. This reflects the fact that populated places further
away from their eventual destination of Srinagarind Hospital must to travel longer to reach the
hospital. This represents a much smoother distribution when compared to the current health
access (see Figure 14), where high travel times exist in small pockets or greatly increase over
short distances. The two strategies generate visually the same results which indicate little
differences in outcomes for individuals.
39

Figure 16 Comparison of population health access between different policy implementation
strategies

Figure 17 shows the travel time to care distributions for the two widespread analyses
compared to results from Section 4.1 in a histogram. In both cases we see the improvements in
distributions. Additionally, for the two widespread strategies we can see a peak at the two hour
mark, which is more pronounced in the bottom frame. This may be due to the fact that since a
diagnosis can occur relatively quickly, most of the travel time being shown is a result from the
distance traveled to reach Srinagarind Hospital.

40

Figure 17 Histogram comparison between different health access policy implementation
strategies

4.3 Optimizing Health Access to Cardiac Care
This section describes the results from the five resource limited implementation strategies
which alter the amount of additional POCT tests and whether the existing resources are
rearranged. Table 12 details the results of the limited-resource evaluations. When compared to
current access all strategies indicate improvement compared to current access. Each strategy
41

offers a steady decrease in travel time the more POCT were available to be integrated into the
health network.

Table 12 Travel time comparison between different low-resource implementation
strategies. Numbers in parentheses show standard deviation.

Health Access

Current

Rearrange
Current
Position
with 5 More
Current
Position with
10 More
Rearrange
with 5
More
Rearrange
with 10
More
Average Travel
from Population
to Diagnosis
49.9
(25.5)
30.6
(15.3)
26.8
(13.1)
21.8
(11.3)
23.3
(12.0)
19.6
(10.6)
Decrease
compared to
Current
N/A 38.7% 46.3% 56.3% 53.3% 60.7%
Average Travel
from Diagnosis
to Care
168.6
(37.1)
150.4
(41.4)
151.4
(39.5)
151.9
(39.9)
149.6
(37.8)
153.3
(38.1)
Decrease
compared to
Current
N/A 10.8% 10.2% 9.9% 11.3% 9.1%
Average Travel
Time from
population to
care
218.5
(39.5)
181
(46.1)
178.2
(42.7)
173.7
(41.9)
172.8
(40.6)
172.9
(39.8)
Decrease
compared to
Current
N/A 16.7% 18.4% 20.5% 20.9% 20.9%

The first analysis, rearranges existing POCT resources to better support the population in
the study area, and resulted in a decrease of 16.7 percent when compared to current health
access. The second and third analysis involves keeping the current POCT where they are and
adding an additional five and ten POCT tests strategically to the health network. These analyses
decreased travel time from populated places to care by 16.7 percent and 18.4 percent
respectively. This puts into perspective that the effect of adding five additional tests in the
second strategy resulted in only a decrease of 1.7 percent, which may not offer enough of a
benefit when compared to the first strategy.
42

In the last two analyses, where existing resources were rearranged along with five and
ten additional POCT, the overall effect on health access remains the same. When comparing
these two strategies the difference between time to diagnosis decreased by 3.7 minutes. However
this decrease was negated by the difference in travel from diagnosis to care, where having more
tests in the health network actually increased travel time. Thus adding the five extra tests does
not warrant the costs of adopting them, since overall it does not add any benefit.
Figure 18 illustrates the travel time from population places to care comparisons from the
five analyses. Much like the results from the widespread analyses, faster travel times start lowest
in the western part of Sakhon Nakhon and increase in magnitude as they spread out. This figure
also demonstrates how it is difficult to visually compare the results of the strategies.
To better compare the strategies, a hot spot analysis using the Getis-Ord G* statistic was
implemented using the travel times from population places to care. These magnitudes were
compared at confidence levels of 90 percent (z-score of 1.65), 95 percent (z-score of 1.96), and
99 percent (z-score of 2.58) to visualize localized hot and cold spots. The results of the hot spot
analysis are shown in Figure 19. Hot spots become smaller as more POCT were implemented
indicating the efficiency of the transportation from diagnosis and care.
Figure 20 shows the travel time comparisons to current health access for corresponding
analyses. Histograms confirm that all strategies offer a benefit over the current implementation
strategy. As with the widespread implementation strategy peaks occur near the two hour mark in
all except where existing resources were rearranged and ten additional tests were integrated.
43

Figure 18 Travel time visual comparison between different low-resource implementation strategies
44

Figure 19 Travel time hot spot analysis for visual comparison between different low-resource implementation strategies

45

Figure 20 Histogram comparison between different low-resource implementation strategies
46

CHAPTER 5: DISCUSSION
The following chapter discusses the data used, critically evaluates the health access model, and
discusses how the model could enable public health decisions.
5.1 Data Sources
One of the goals of this project was to provide useable outcomes using data sets that were
readily available. This section discusses issues encountered with the data, their effect on the
analysis, and other data sources which could be used for better outcomes.
5.1.1 Population
Location where individuals originate is a major input for this model. In this study
populated places from OSM were used. It is apparent that this data source may not be
appropriate for two reasons. The first is that there is no means to judge the accuracy or the
completeness of this of dataset. The second reason is that the populated places may not represent
the actual originating locations for individuals.
To overcome incompleteness or inaccuracy of OSM data, census totals can be used.
Census aggregations may provide a better and more standardized way of obtaining origins of
travel. However, the data collection timeframe and spatial aggregation may not be of appropriate
scale. Census data are usually aggregated by population density. Sparsely populated areas are
usually aggregated to larger area then dense populations. This may result in a poor understanding
of health access due to generalization of population patterns.
One way to better represent stating locations of individual in this model is to use activity
space. Activity spaces are geographically defined zones centered around the home in which
everyday life unfolds (Cromley and McLafferty 2012). Activity spaces emerge from the social
47

demand from the individual’s community or household. This is a complex spatial concept which
may be difficult if not impossible to incorporate into model as it exists right now. However, it is
clear that a single point cannot be used to accurately represent a starting location for an
individual. Activity spaces offer a better understanding of where individuals will be when they
need to access their health network.
5.1.2 Roads
A major aspect of this analysis is the data used in the road network, since it determines
how the model calculates travel time and distance. The model demonstrated that variations in
travel speeds can drastically affect health access (Table 7). The study also demonstrated that
large areas of are poorly represented by OSM due to lack of data (Figure 7).
One way to alleviate this problem is to adopt either nationally recognized data sets from
the government, such as Topologically Integrated Geographic Encoding and Referencing
(TIGER) roads for the United States, or commercial datasets, such as Tom Tom® and HERE®.
These datasets may provide a better understanding of the road networks without gaps in data.
The downside is that robust public data sets may not exist in certain countries. Additionally,
commercial data sets may not be available in developing countries.
5.2 Critically Evaluating the Model
This model assumes that an individual chooses to take the quickest path to their nearest
diagnostic resource and care. There are several different ways in which an individual may
realistically divert from this assumption. First, a patient who self-identifies as critically ill may
decide to travel in a more direct path knowing their ending destination is will be Srinagarind
Hospital. Second, a patient may not have the knowledge of or does not trust certain roads or
health resource facilities and thus alters their path, from a more optimal one.
48

These diversions from the modeled paths represent the concept of SWN, where
individual paths through the network directly reflect on the cultural or social pressures that exist.
A way to accurately include this into the model is to survey population or health professionals to
better understand their attitudes towards their health network. This can help better understand
how the population makes decisions on how they utilize their health networks.
Another limitation of this model is that the quickest path does not necessarily indicate the
most efficient path to the end goal: care. This is due to the fact that the model only optimized the
time to diagnosis. As identified Section 4.1, traveling to the closest facility with appropriate
diagnosis may not be the most effective overall care path. Another way to preposition supplies is
to optimize travel from populated places to not only a quick diagnosis but also ensuring access to
care is optimized as well.
This model was created to understand a SWN in a rural low-resource location where
limitations in technologies are well known. It will be interesting to see how the model behaves in
more developed countries like the USA or Europe. In these locations, health access may already
be optimized, in which case this model may only suggest subtle improvements or that no
improvements are needed.
5.3 How this model fits into future public health decision
Health costs have consistently risen over the last few years which partially can be
associated with technology development. Additionally, integration of new technology is often
slow, difficult, and expensive. One of the reasons is that health networks may not have the means
to understand the benefits of adopting technologies or relate those benefits to the costs of
adoption (Price and John 2012).
49

The hope for this model is that it will fit into future decision making for all health
networks by evaluating care paths in spatial framework. A spatial care path
TM
, first defined by
investigators Dr. Gerald Kost and William Ferguson as the most efficient route available to
individual patients within health networks, uses geospatial information to improve decision-
making and reduce costs (Kost, Ferguson, and Kost forthcoming). The expectation for the model
used in this study is for it to evolve to be able to understand and implement the spatial care
path
TM
concept within health networks.
The model, in its current condition, is not ready to be used to make decisions. Although it
can be a useful tool to understand the general spatial benefit and umbrella strategies of
implementing POCT, the processes modeled, data used, and the fact that it has not been validated
means it is not sophisticated enough to be used in the real world.
50

CHAPTER 6: CONCLUSIONS
George E.P. Box wrote that “all models are wrong; the practical question is how wrong do they
have to be to not be useful” (Box and Draper 1987, 74). This is relevant because spatial models
can be defined as a “simplified representation of a system under study, which can be used to
explore, to understand better or to predict the behavior of the system it represents” (O’Sullivan
and Perry 2013, 3). Because spatial models are simplified representation of reality they involve
making assumptions which inherently bias the model. This bias must be understood as to how it
alters the conclusions we can reliably draw to make decisions in reality.
This model was created to understand the role of diagnostic resources within a health
network. It did this by first performing a sensitivity analysis which allowed us to put into
perspective how the model parameters could mislead the analysis. The second step was to
understand how it calculates current health access. The third step is modelling alternative
scenarios which can be compared to the current health access, helping make decisions on the best
implementation strategies. To take these decisions into the real world we must understand how
these benefits relate to the costs associated with the technology integration and how the modeled
benefits deviate from reality.
While this project has answered many questions on how the model behaves using
available data, it has raised many questions as well. How do we account for the incomplete OSM
data? How does data vary from country to country? How do we better represent the starting
locations for individual within health networks? How does this model change when evaluating
urban environments or in developed countries? These questions will be answered as the model
grows and matures. The final challenge will be to validate the model to prove that a structured
spatial analysis of existing SWN will facilitate decisions improving health networks.
51

Despite its preliminary nature, this project has produced two significant results. The first
is that this project has helped demonstrate how adopting POCT into health networks can
streamline decision making at the point of need. The second is that this model has shown
promise in being able to quantify the effects of implementation strategies by relating the benefits
of adopting POCT to the costs of integrating them. The hope for this spatial model is that it can
be used to make effective and efficient decisions that will improve individual’s health access
within health networks.
52

REFERENCES
Blanchet K., P. James. 2012. How to do (or not to do)… a social network analysis in health
systems research. Health Policy and Planning. 27: 438-446.
Bolstad P. 2012. GIS Fundamentals: A First Text on Geographic Information Systems. White
Bear Lake, MN: Eider Press.
Box, G. E. P., and N. R. Draper. 1987. Empirical Model Building and Response Surfaces. New
York, NY: John Wiley & Sons.
Brabyn L. and C. Skelly. 2002. Modeling population access to New Zealand public hospitals.
International Journal of Health Geographic. 1 (3): 1-9.
Chambers D., P. Wilson, C. Thompson, and M. Harden. 2012. Social Network Analysis in
Healthcare Settings: A Systematic Scoping Review. Public Library of Science. 7 (8).
Clark R. A., N. Coffee, D. Turner, K. A. Eckert, D. van Gaans, D. Wilkinson, S. Stewart, and A.
M. Tonking. 2012. Application of Geographic Modeling Techniques to Quantify Spatial
Access to Health Services Before and After an Acute Cardiac Event. Health Services and
Outcomes Research. 125: 2006-2014.
Coffee N., D. Turner, R. A. Clark, K. Eckert, D. Coombe, G. Hugo, D. van Gaam, D.
Wilkinson, S. Stewart, and A. A. Tonkin. 2012. Measuring accessibility to cardiac
services using geographic information systems. Applied Geography. 34: 445-455.
Cromley E., and S. McLafferty. 2011. GIS and Public Health. New York, NY: The Guildford
Press.
Curtis C. M., R. F. Louie, J. H. Vy, W. J. Ferguson, M. Lam, A. T. Truong, M. J. Rust, and G. J.
Kost. 2013. Innovations In point-of-care testing for enhanced United States disaster
caches. American Journal of Disaster Medicine. 8 (3): 181-204.
53

Delamater P, J., P. Messina, A. M. Shortridge, and S. C. Grady. 2012. Measuring geographic
access to healthcare: a raster and vector based approach. International Journal of Health
Geographics. 11 (15): 1-18.
Garcia P. J., C. P. Carcamo, M. Chiappe, M. Valderrama, S. La Rosa, K. K. Holmes, D. C. W.
Mabey, and R. W. Peeling. 2013. Rapid Syphilis Tests as Catalysts for Health Systems
Strengthening: A Case Study from Peru. Public Library of Science. 8 (6).
Jordan H., P. Roderick, D. Martin, and S. Barnett. 2004. Distance, rurality and the need for care:
access to health services in South West England. International Journal of Health
Geographics. 3 (21): 1-9.
Kost G. J. 2012. Theory, Principles, and Practice of Optimizing Point-of-Care Small-World
Networks. Point of Care. 11 (2): 96-101.
Kost G. J., W. J. Ferguson, L. E. Kost. 2014. Principles of point of care culture, the spatial care
path,
TM
and enabling community and global resilience. The Journal of the International
Federation of Clinical Chemistry and Laboratory Medicine. 25(2): 7-25 (Forthcoming).
Kost G. J., L. E. Kost, A. Suwanyanyuen, S. K. Cheema, S. Sumner, J. Yu, and R. F. Louie.
2010. Emergency Cardiac Biomarkers and Point-of-Care Testing: Optimization Acute
Coronary Syndrome Care Using Small-World Networks in Rural Settings. Point of
Care. 9 (2): 53-64.
Kost G. J., N. K. Tran, and R. F. Louie. 2002. Point-of-care testing: principles, practice, and
critical emergency-disaster medicine. In: Encyclopedia of Analytical Chemistry. ed. R.A.
Meyers. Hoboken, NJ: John Wiley & Sons Ltd.
54

Kost G. J., J. N. Yu, and N. K. Tran. 2010. Theory, Principles, and Practice of Optimizing
Point-of-Care Small-World Networks for Emergency and Disaster Care. Transactions
on Biomedical Engineering. 11 (2): 96-101.
Law, G. 2014. Provinces of Thailand. http://www.statoids.com/uth.html (last accessed August
26, 2014).
Lewandrowski E. L. and K. Lewandrowski. 2013. Implementing point-of-care testing to
improve outcomes. Journal of Hospital Administration. 2 (2): 125-132.
Luke, D. A., K. S. Stamatakis. 2012. Systems science methods in public health: dynamics,
networks, and agents. Annual Review of Public Health; 33: 357-76.
Mabey D. C., K. A. Sollis, H. A. Kelly, A. S. Benzaken, E. Bitarakwate, J. Changalucha, X. S.
Chen, Y. P. Yin, P. J. Garcia, S. Strasser, N. Chintu, T. Pang, F. Terris-Prestholt, S.
Sweeney, and R. W. Peeling. 2012. Point-of-care test to strengthen health systems and
save newborn lives: The Case of Syphilis. Public Library of Science. 9 (6).
MapMagic Thailand 2013. Version 1.0.1. THiNKNET Co., Ltd. Bangrak, Bangkok.
Nykiforuk C. I. J., L. M. Flaman. 2011. Geographic Information Systems (GIS) for Health
Promotion and Public Health: A Review. Health Promotion Practice; 12 (1): 63-73.
O’Sullivan D., G. L. W. Perry. 2013. Spatial Simulation: Exploring Pattern and Process.
Oxford UK: Wiley-Blackwell.
Owen K. K., E. J. Obregon, and K. H. Jacobsen. 2010. A geographic analysis of access to health
services in rural Guatemala. International Health. 2: 143-149.
Peeling R. W. and D. Mabey. 2010. Point-of-care tests for diagnosing infections in the
developing world. Clinical Microbiology and Infection. 16 (8): 1062-1069.
55

Price P. P, A. S. John. Point-of-Care Testing: Making Innovation Work for Patient-Centered
Care. Washington DC: AACC Press.
Ranisinghe I., F. Turbull, A. Tonkin, R. A. Clark, N. Coffee, and D. Brieger. 2012.
Comparative Effectiveness of Population Interventions to Improve Access to
Reperfusion for ST-Segment-Elevation Myocardial Infarction in Australia. Circulations
Cardiovascular Quality and Outcomes. 5: 429-436.
Schuurman N., R. S. Fiedler, S. C. W. Grzybowski, and D. Grund. 2006. Defining rational
hospital catchments for non-urban areas based on travel-time. International Journal of
Health Geographics. 5 (43): 1-11.
Watts D. J. and S. H. Strogatz. 1998. Collective dynamics of ‘small-world’ networks. Nature.
393: 440-443.

Abstract (if available)

Abstract Rapid and accurate diagnoses are important because they drive evidence-based care in health systems. Point-of-care technologies (POCT) can aid in diagnosis by bringing advanced technologies out of hospital or clinical settings and closer to the patients. Health networks are constrained by natural connectivity in the interactions between geography of resources and social forces. Using a geographic information system (GIS) we can understand how populations utilize their health networks, visualize their inefficiencies, and model alternatives. This project focuses on cardiac care resource in rural Isaan, Thailand. A health access model was created using ArcGIS Network Analyst 10.1 from data representing aggregated population, roads, health resource facilities, and diagnostic technologies. This model was used to quantify current cardiac health access and improve upon that access using both widespread and resource limited strategies. Sensitivity analysis revealed that altering travel speeds of roads has a large effect on the calculation of health access. Results indicated that having diagnostic technologies closer to population allowed the streamlining of care paths. The model allowed for comparison of the effectiveness of the implementation strategies. This model was created to help put the benefit of adopting POCT within health networks within perspective. Additionally it can help evaluate these alternatives diagnostic placement strategies as compared to the current health access and evaluate the relative costs and benefits.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Use of GIS for analysis of community health worker patient registries from Chongwe district, a rural low-resource setting, in Lusaka Province, Zambia

PDF

Address points and a maser address file: improving efficiency in the city of Chino

PDF

Spread global, start local: modeling endemic socio-spatial influence networks

PDF

Integrating spatial visualization to improve public health understanding and communication

PDF

Modeling burn probability: a Maxent approach to estimating California's wildfire potential

PDF

Congestion effects on arterials as a result of incidents on nearby freeway: When should you get off the highway?

PDF

Evaluating predator prey dynamics and site utilization patterns of golden eagles using resource selection modeling and spatiotemporal pattern mining

PDF

Preparing for immigration reform: a spatial analysis of unauthorized immigrants

PDF

Modeling historic structure preservation candidacy on Fort Ord

PDF

The spatial effect of AB 109 (Public Safety Realignment) on crime rates in San Diego County

PDF

A threat-based least-cost path decision support model for national security resource allocation along the US-Mexico border

PDF

3D visualization models as a tool for reconstructing the historical landscape of the Ballona Creek watershed

PDF

Wrong way driving in San Antonio, Texas: a transportation route study using Network Analyst

PDF

Mapping uniformity of park access using cadastral data within Network Analyst in Wake County, NC

PDF

Using pattern oriented modeling to design and validate spatial models: a case study in agent-based modeling

PDF

Social media canvassing using Twitter and Web GIS to aid in solving crime

PDF

Generating bicyclist counts using volunteered and professional geographic information through a mobile application

PDF

Surface representations of rainfall at small extents: a study of rainfall mapping based on volunteered geographic information in Kona, Hawaii

PDF

A Web GIS application for airport pavement management

PDF

The role of GIS in asset management: County of Kauai Department of Parks and Recreation a need for an asset management program

Asset Metadata

Creator Ferguson, William John (author)

Core Title Modeling patient access to point-of-care diagnostic resources in a healthcare small-world network in rural Isaan, Thailand

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 09/09/2014

Defense Date 08/21/2014

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag diagnostic resources,health access,network analysis,OAI-PMH Harvest,spatial modeling

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kemp, Karen K. (committee chair), Chiang, Yao-Yi (committee member), Ruddell, Darren M. (committee member)

Creator Email wjfergus@gmail.com,wjfergus@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-471617

Unique identifier UC11286842

Identifier etd-FergusonWi-2901.pdf (filename),usctheses-c3-471617 (legacy record id)

Legacy Identifier etd-FergusonWi-2901.pdf

Dmrecord 471617

Document Type Thesis

Format application/pdf (imt)

Rights Ferguson, William John

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA